OpenAI co-founder and president Greg Brockman said at an event that AI models are becoming so advanced and multi-faceted that engineers are struggling to monitor and control them.
Brockman, while speaking at a private OpenAI media roundtable in New York City this week, addressed a recent incident where one of the company’s AI models broke out of what was supposed to be a secure sandbox and hacked into Hugging Face, an online platform for AI learning.
The model believed Hugging Face had solutions that could allow it to cheat on an assessment it had been given, according to OpenAI.
“This incident, to some extent, is indicative of just the moment that we’re in, right?” Brockman said, according to Fortune. “Sometimes it’s hard to lose track of any one dimension that [the AI models are] actually very capable at.”
ANTHROPIC CALLS FOR INDUSTRY-WIDE AI SAFETY STANDARDS TO KEEP MODELS FROM WREAKING HAVOC
Brockman said the model’s ability to target Hugging Face underscored how good OpenAI’s products are at cybersecurity tasks.
“Can we be in a world where defenders are able to spend 10 times as much compute defending and making sure every single piece of software that we have is fully secure relative to anyone else?” Brockman said, according to Fortune.
Brockman said OpenAI is taking the breach “very seriously.”
However, OpenAI has also used the event to pitch its own models for their cybersecurity prowess. As part of that, the company has asked potential clients to apply for a “trusted partner” status to get access to these models.
GOOGLE LAUNCHES STUDY OF MILLIONS OF AI CHATS TO UNDERSTAND HOW PEOPLE USE ARTIFICIAL INTELLIGENCE
“We encourage other defenders to apply for trusted access and experiment with these models now to translate these capabilities into better prevention, faster detection, and more effective incident response,” OpenAI wrote in a blog post.
While at the event in New York, Brockman also addressed questions about the Trump administration considering a ban on Chinese-made AI models. Axios reported on this potential policy earlier this week.
“AI is something that is very important to democratize,” Brockman said, before saying having access to more models is a good thing. He did not explicitly say whether he’d support a ban on AI model made by companies based in China.
Per FEC filings, Brockman has become one of the biggest donors to President Donald Trump’s political movement, last year giving $12.5 million to MAGA Inc., a Trump-aligned Super PAC.
LOCK DOWN YOUR CHATGPT ACCOUNT BEFORE THE NEXT AI ATTACK
Brockman said he has not been in conversation with any Trump officials about a blanket ban on Chinese AI.
“For any model, it’s not really about who creates it,” he said. “How do you evaluate a model? How do you think about its safety? How do you think about its use cases? How do you understand its alignment?”
The Trump administration is weighing a ban after Beijing-based Moonshot AI released its Kimi K3 model, Axios reported.
The model may be able to equal the performance of the best models from American firms OpenAI and Anthropic, but at a fraction of the cost.
Michael Kratsios, a science advisor to the president, publicly accused Moonshot of using American AI companies’ models to train their own, thereby infringing on their intellectual property rights.
“We have information that Moonshot AI distilled Anthropic’s Fable for the development of its K3 model,” Kratsios wrote in a social media post on Wednesday. “Large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology and undermining American research is unacceptable.”
The U.S. government’s concerns about Chinese AI models is not necessarily shared by everyone in the AI industry, including Brockman and Nvidia CEO Jensen Huang.
Huang this week called the latest AI models to come out of China “excellent” and said that they “should be used.”
Read the full article here
